Anthropic has released Claude Opus 4.5, and the company describes it as their smartest and safest AI so far.
If you’re not an engineer, here are the upgrades in clear, simple terms — all based on Anthropic’s official announcement.
1. It Understands Problems Better
Claude Opus 4.5 is much better at figuring things out on its own.
Anthropic says it can:
- →Understand tricky questions more clearly
- →Handle ambiguity
- →Work through complicated tasks without getting lost
In their internal tests, it even solved difficult challenges faster and more accurately than before.
2. It’s More Helpful With Everyday Work
The new version is stronger at tasks many people do daily:
- →Working with slides and spreadsheets
- →Helping with research
- →Managing long conversations without forgetting earlier parts
- →Using computers more effectively (like navigating apps or browser tabs)
For example, Anthropic says that in the Claude app, you can now have long, uninterrupted chats, because the model automatically summarises what happened earlier.
Opus 4.5 is designed to be better at acting like a “digital assistant”:
- →It can follow multi-step instructions
- →It can use tools like Excel or browser extensions more reliably
- →It can coordinate several smaller “sub-agents” to complete bigger tasks
In other words, it’s better at doing work for you, not just answering questions.
Anthropic shows that this leads to big improvements in long, complex tasks.
4. It’s Safer and Harder to Trick
Anthropic highlights safety as a major focus.
According to their benchmarks:
- →Opus 4.5 is harder to manipulate with harmful or deceptive prompts
- →It shows fewer risky behaviours
- →It’s their most aligned model to date
They also show charts where Opus 4.5 resists hacking-style “prompt attacks” better than other major models.
5. It Uses Fewer Tokens (Which Means Lower Cost and Faster Work)
For people using the API or advanced settings:
Anthropic added a new control that lets you choose how much the model should “think.”
At medium level, Opus 4.5 matches the performance of the previous model while using 76% fewer tokens.
At high level, it can outperform it while using about half the tokens.
This means:
- →Faster responses
- →Less cost for the same (or better) quality
- →More efficient long tasks
6. Improvements Across the Whole Claude Ecosystem
Anthropic also updated their apps and tools:
Claude Code (for developers)
- →Now creates clearer plans before running
- →Available on desktop, so you can run multiple sessions
Claude App
- →Long conversations no longer get cut off
- →Better with documents, spreadsheets, and research
Claude for Chrome
- →Available for all Max users
Claude for Excel
- →Beta access expanded to more users
All of these features rely on Opus 4.5’s improved ability to manage long tasks and use computers effectively.
7. Price and Availability
Anthropic lists pricing at:
- →$5 per million input tokens
- →$25 per million output tokens
And the model is available in:
- →Claude apps
- →Claude API
- →All major cloud providers
This summary is based entirely on Anthropic’s official announcement.
Source: https://www.anthropic.com/news/claude-opus-4-5
